Following is a brief summary of activities and progress of the University of Michigan Digital Library (NSF-UMDL) project from December, 1994 through April, 1995. The project is continuing to move forward, and is ahead of schedule in fulfilling some of the scheduled milestones and tasks for the year.
In addition to the development and dissemination of the NSF-UMDL, the project team participated in the first annual site review in February, and participated extensively in the recent all- project meeting at the University of Illinois, Urbana-Champaign. Project teams continue to meet on a weekly basis, including the operating committee which guides the overall management of the project. PERT charts including a record of activities and tasks for the coming year are maintained. Two new members will be joining the project:
The architecture group implemented and demonstrated the first prototype system. This system, which follows the Architectural Specification Version 1.0, is organized around the following agents:
The agents are implemented as a set of procedures (APIs), specific to each type of agent. The APIs can be used by third parties create their own UMDL-compliant agents. The APIs are invoked as dictated by the pragmatics of the interagent communication language (currently KQML). Moreover, more complex, context-dependent procedural responses by task planning agents are represented and incrementally invoked within a procedural reasoning architecture that has been used for implementing the agents.
Two of the major objectives of the architecture are scalability and support for heterogeneous agents (collections and user interfaces). The current prototype contains about 50 CIAs, and can gracefully accept many more (they can also be gracefully removed); the same holds for UIAs. Thus, the system has scaled well in the initial trials (we've had upwards of 100 agents running in the UMDL). Furthermore, the collections use a variety of search engines and document types.
Our object this calendar year is to provide better search support and other services (e.g., subscription and notification services). Toward this end, we have begin to architect and implement three new classes of agents:
These specifications of this agents will be published at the end of Q3 '95.
We will also be extending the capabilities of the task planning mediator agents to support representing procedures that involve the interweaved activities of several agents. A task planner responsible for accomplishing a more difficult goal can retrieve such a procedure and (possibly with the assistance of other mediators) recruit agents to perform portions of the procedure in a distributed manner, so that a team of agents can accomplish a goal that exceeds the abilities of the agents individually.
The methodologies being used to evaluate the user interface are focused-group interviews (with prospective users such as high school student science majors and non majors, media specialists, public library users), interface literature review, review of interfaces of existing systems (e.g., encyclopedias, games, reference works), walkthroughs that include observation, in-search and post-search interviews, protocol analysis (including walkthroughs that use video or print- on-paper mock-ups of system activity and screen displays in lieu of actual system use), and transaction log analyses. Plans are underway to begin conducting reviews of the relevant professional literature and existing interface, and focused-group interviews. These findings will be integrated into the development of a prototype UIA, from which user reactions and responses will refine the prototype UIA.
Figure 1. Flowchart of UIA research and development efforts
Objectives and research questions connected with these methodologies are :
Specific UIA features we are considering for development in 1995 are:
The Collection Interface Agents (CIA) working group has been involved in discussions and development in three major areas. The first involves the design of methods for adding new collections and search engines to the UMDL. Procedures and criteria for evaluating collections and engines for inclusion in the system are being outlined. These criteria are intended to favor factors such as current market share, adherence to open architecture standards, and utility to end users of UMDL. The team is examining mechanisms for searching both the Conspectus and the individual collections which are mounted under separate CIAs.
Exploration of specific types of searches which will be issued to the Conspectus, and the software components necessary to conduct them, are being examined. Capabilities for searching not only by the broad topics already found in the Conspectus, but by other " collection level" elements such as author and more specific subjects are being developed. Finally, we are developing procedures for identifying the best search engines for specific collections or collection types (an evaluation function) and we are developing an experimental component aimed at testing possible new search techniques on selected collections.
A specification of information goods and services is being developed. Substantial effort has been devoted to the design of a vocabulary for describing information services available in the digital library. This language will be used to specify agent capabilities and as the basis for negotiation of various goods and services in the NSF-UMDL. As part of this activity a framework for describing intellectual-property rights, a critical feature of the services available in the digital library, has been developed.
The initial system for economic resource allocation in the NSF-UMDL, based on the concept of market facilitators (auction agents), has been designed. Basic protocols for communicating with auction agents have been worked out, and their behavior has been specified in a draft design document. This working design document, along with others, can be found on IPE WWW page (http://http2.sils.umich.edu/UMDL/ipe/IPE-index.html).
Several collections have become available through the NSF-UMDL. UMI has delivered 43 journals including both the bit mapped images and the citation data. This content is available through the NSF-UMDL. Elsevier began delivering initial bit mapped images of 54 journals in March, 1995. In addition to these images, SGML samples (to work with the DTD's) have been received. The McGraw-Hill Encyclopedia of Science and Technology was received and added to the collection in April, 1995. Negotiations for additional content are underway with McGraw-Hill. The American Chemical Society provided 5 years of the Encyclopedia of Environmental Technology in SGML format. This content is being added to the collection. Relevant Internet resources have been identified, described in the Conspectus, and are available through the NSF-UMDL. The project has obtained commitments from Elsevier for 10 book series and reference works, the Windows on the World Project for NASA data, and Encyclopedia Britannica Educational Corporation for the video series Galactic Encyclopedia and Problems of Conservation.
Negotiations for additional content are underway with several publications. A list of publishers and content is outlined below.
Groliers Americana Encyclopedia (SGML)
Cambridge University Press journal and monograph titles
(relevant to subject domain)
InfoSoft dictionary, thesaurus, almanac
(SGML)
Ebsco high school level journals
University of Chicago Press access to Astrophysical Journal
Association for Computing Machinery exploring conversion of backfiles
Deployment of the NSF-UMDL and baseline evaluation is underway at several sites. Building infrastructure and culture in the test sites in Ann Arbor has been a complex task. However, infrastructure is completed at both Pioneer and Community high schools, and is nearly complete at the Ann Arbor Public Library. The high school teachers, media specialists, and public librarians have received both Internet and NSF-UMDL training. Integration of the NSF-UMDL into the school curriculum was begun during the winter semester.
At the University of Michigan, discussions are underway to incorporate the Conspectus registry into the UM Information Gateway (institutional homepage and framework for UM campus information resources), replacing the existing database and classification scheme. In addition, the project is pursuing an " advanced interface" option within the Gateway which would launch the NSF-UMDL UIA. Resources have been committed to enhance the existing FTL search engine to incorporate a Z39.50 gateway and allow linkages/communication between the campus OPAC and other bibliographic resources. The Digital Library Initiatives electronic newsletter will be launched in spring 1995 to highlight NSF project and other activity on campus.
Project team members are continuing to build collaborative relationships with research and equipment partners. Team members are engaging in ongoing meetings with Apple, IBM, Xerox PARC, Bellcore and other partners to explore further opportunities for technical interactions. In addition, plans for the first External Advisory Committee meeting are underway.
Comments or questions may be sent to: UMDL.INFO@umich.edu